Predicting PACS loading and performance metrics using Monte Carlo and queuing methods

Determining the performance bottleneck of a PACS system is a challenging task. System performance is dependent on several variables such as the workstation, network, servers, type of data, and different loading conditions. This makes planning difficult to ensure the system capacity will deliver fast access to images throughout the enterprise of a hospital even during rush periods. The rules of thumb that most vendors use for the number of workstations per server are based upon heuristic experience and may not apply from institution to institution where usage and infrastructures are different. Rules of thumb can be problematic and usually cannot predict the impact when new technology is introduced like Gigabit Ethernet or distributed architectures. We have developed a Monte Carlo Model in an attempt to develop a more accurate model to predict loading on a system at peak "rush hour" times. The focus of the model was on user metrics of performance such as the latency and throughput of images to their workstation. Analysis demonstrates that "traffic jams" can occur and dissipate in a matter of minutes and be relatively irreproducible to the PACS administrator.